Using AI in Australian Public Sector workplace investigations

The rapid evolution of artificial intelligence (AI) tools presents both risks and opportunities for the conduct of internal workplace investigations across the Australian Public Sector.

We provide guidance to Commonwealth agencies about the effective use of AI in supporting internal workplace investigations into concerns such as allegations of bullying, harassment, sexual harassment and similar misconduct.

The rise of AI-supported workplace complaints

Across all sectors, employers are facing rising workplace complaints. Anecdotally, workplace grievances are increasing in both frequency and complexity. In a recent case before the Fair Work Commission, Deputy President upheld the dismissal of a serial workplace complainer who, with the help of AI, had "become ungovernable due to his voluminous and obstructive communications".

These trends are also reflected in the volume of claims being initiated in jurisdictions like the Fair Work Commission, which has reported a 70% increase in claims over the three years to date.

Increasing complaints are attributable, at least in part, to the rise and accessibility of AI tools which are supporting employees to raise complaints - often leading to complex and protracted internal workplace investigations.

Reducing psychosocial risk through timely investigations

Best-practice requires workplace investigations to be conducted promptly and without unreasonable delay. By harnessing AI tools safely and effectively, there is an opportunity for Commonwealth agencies to conduct their workplace investigations more efficiently, thereby reducing the risks associated with delay.

Similar risks were highlighted recently when SafeWork NSW issued notices to the NSW Department of Education for its failure to ensure that workplace investigations into misconduct and performance concerns were conducted in a timely and expeditious manner, exposing employees to psychosocial hazards.

While delay is sometimes unavoidable (for example, where witnesses take unexpected leave), other delays can be reduced by using AI to support investigation processes.

Key principles for using AI in workplace investigations

Consistent with best-practice, the APS Values, Code of Conduct and Policy for the responsible use of AI in government when using AI to support a workplace investigation, investigators should be mindful of the following principles:

• any use of AI by Commonwealth agencies must be in-scope and consistent with the agency's own policies, procedures and appliable law
• investigations must be conducted with care and diligence. Human investigators remain accountable for their investigation processes, reasoning, findings and outcomes
• maintain confidentiality and privacy. In the course of an investigation, classified, confidential or sensitive information should never be disclosed to open-source AI tools, nor any AI tool that is not approved for use by the relevant Commonwealth agency for that purpose
• always verify AI output before relying on it in any workplace investigation
• practising lawyers who are conducting workplace investigations have additional, professional obligations to exercise their own forensic judgement when providing legal advice in connection with any workplace investigation
• investigators should keep accurate records of how and when AI has been used to support an investigation process to ensure transparency.

Practical uses of AI in workplace investigations

Some examples of the effective use of AI to support an investigation may include the following.

• investigation planning - AI can be an effective tool for planning. For example, by prompting a permitted AI tool to review the agency's policies, processes or enterprise agreements and asking it to prepare a detailed investigation plan. Mapping-out the required steps and timeframes can help with the smooth running of an investigation and ensure compliance
• records of interview - AI tools can save time by preparing records of witness interviews. However, investigators must ensure that appropriate notification or consent has been provided by the witness before recording and transcribing an interview. Records of interview prepared with AI assistance should aways be verified by the investigator and, in most cases, by the witness themself
• preparing chronologies, filtering and sorting evidence - AI tools can save time by organising data, filtering information, and looking for patterns. For example, organising documents chronologically, filtering emails, or searching for specific patterns. Investigators should think critically about the prompts that are given to AI tools and always independently check the accuracy of the output before relying on it in an investigation
• content generation - Generative AI can save time by generating first drafts of procedural documents used in investigations, such as written communications with witnesses, or letters asking for information.

However, investigators should not outsource critical thinking to AI. Assessing credibility, weighing up the evidence, making findings and ensuring procedural fairness are the responsibilities of the investigator, not AI tools. Critical aspects of investigation reports should, therefore, be product of the investigator's independent thought, not AI.

Ultimately, any APS employee who is conducting a workplace investigation for a Commonwealth agency remains accountable for the process, reasoning and the findings of their investigation. When used properly, AI can support the investigator, not replace them.

Read more about workplace surveillance in our related article.